How to get AI to build an app for you (no code, five steps)

Short answer: describe the problem in plain language instead of describing the software, pick where the thing will live, give it one easy way to put information in, let the AI do the part that needs understanding, then change it after you have used it for a couple of days. You do not need to write code. You need to write a good brief.

Here is what that looked like for me.

The problem I actually had

I am going to London for a week in December. For a month beforehand, every time someone sent me a reel of a bar or a market or a restaurant, I saved it. By the time I looked properly there were dozens of posts sitting in Instagram with no names, no addresses and no order to them.

I was going to forget most of it. Saved posts are where good ideas go to die.

So instead of building a spreadsheet, I asked Claude to build me something.

What it does now

I text a phone number an Instagram link. That is the whole interaction.

A few seconds later the place is in my itinerary with its name, what it is, where it is, and how far it is to walk from my hotel. I can open it and filter the list to cafes, restaurants or Christmas markets, and see them on a map with walking times from where I am staying.

I never told it what any of those places were. It worked that out from the links.

1. Write down the problem, not the solution

One paragraph. What happens now, why it is annoying, and what you want instead.

Resist naming features. If you open with "it needs a database and a map view", you have already shrunk the result down to what you could imagine on the spot. I described a person walking around a city with a phone, not a piece of software.

The quality of what you get back is set by the quality of that description. Vague in, generic out.

2. Decide where it lives before you build anything

A web app, a Notion database, a spreadsheet, a page on your phone. Mine is a small web app because I wanted a map. Yours may not need one.

This decision shapes everything after it, so make it deliberately rather than letting the AI pick for you.

3. Give it one easy way in

Mine is a text message, because texting is something I already do. The way in should be a habit you have, not a new one you have to remember.

This is where most personal tools die. If adding something takes more effort than the old messy way, you will stop within a week.

4. Let the AI do the part that needs understanding

I send a link. Something has to open it, work out that it is a wine bar in Soho, and file it in the right place.

That step is the reason this was not possible a couple of years ago. Everything else is plumbing. When you are deciding whether AI belongs in a job, look for the step that needs judgement or comprehension. That is the part worth handing over.

5. Use it, then change it

I did not specify all of this up front. I used it for two days, noticed I kept wanting to know how far places were from my hotel, and asked for that. Then I wanted filters by category.

Each round took minutes. The second version is always better, because by then you know what you actually wanted.

Do you need to know how to code?

No. I did not write any code for this, and I am not a developer.

What you do need is the ability to describe a process clearly: what triggers it, what happens in the middle, and what good looks like at the end. That is a thinking skill, not a technical one, and it is the same skill that makes handing repeat work to Claude work inside a business.

How long does something like this take?

The first working version took an afternoon. The changes after that took minutes each.

Most of the afternoon was spent describing the problem properly, not waiting for anything to build. That ratio is normal and it surprises people.

It is also the same shape as plenty of business problems. An inbox is a list of things arriving that need sorting and acting on, which is why running your inbox with Claude uses the same five steps. So does building a CRM in Notion.

Why I built this for a holiday, not for work

Most of what I teach is AI inside a business. Operations, inboxes, content, leads. This was not that. I built it for a trip.

But it is the same skill, and that is the point. Spotting that something is repetitive, rules-based and currently manual is the whole game. It does not matter whether the process is invoicing or a list of pubs.

If you want to get better at AI in your business, start with something small and personal where nothing is at stake. You will learn more from one weekend project you actually use than from another course.

How to get Claude to build your first app

Describe the problem properly and Claude Code can have a working first version in front of you in an afternoon, without you writing any code.

Use Claude Code. You are building a real thing that has to keep running: files, a place to live, and something that reacts when you send it information. That is Code, not Cowork. You do not need to know how to code, but you do need to describe a process clearly.

You need a Claude account with Code, somewhere for the thing to live, which you decide in step two, and one paragraph describing what currently goes wrong.

1. Write one paragraph about the problem, not the software. What happens now, why it is annoying, what you want instead.
2. Decide where it lives before you build anything: a web page, a Notion database, a spreadsheet, something on your phone.
3. Pick one way to put information in that is already a habit for you, not a new one to remember.
4. Give Claude the prompt below and let it build the smallest working version.
5. Use it for two days before you ask for a single change.
6. Then change it, one thing at a time, in the same conversation.

I want you to build me something. Read all of this before you start.

The problem: [ONE PARAGRAPH. WHAT HAPPENS NOW, WHY IT IS ANNOYING, WHAT YOU WANT INSTEAD. DO NOT NAME FEATURES.]

Where I want it to live: [WEB PAGE / NOTION / SPREADSHEET / PHONE]
How I want to put things into it: [THE THING YOU ALREADY DO: TEXT A NUMBER, FORWARD AN EMAIL, PASTE A LINK]
What good looks like: [HOW YOU WILL KNOW IT IS WORKING]

The part that needs judgement: [THE STEP THAT NEEDS SOMETHING UNDERSTOOD RATHER THAN MOVED. THIS IS THE BIT WORTH HANDING TO AI.]

Before you build anything:
1. Tell me back what you think I am asking for, in your own words
2. Ask me whatever you still need to know
3. Tell me the simplest version that would still be useful

Then build THAT version only. Do not add features I did not ask for. Show me it working on one real example.

Expect the afternoon to go on describing the problem rather than waiting for anything to build. That ratio is normal and it catches people out. The second version is always better, because by then you know what you actually wanted.

Where to start this week

Pick the thing you keep collecting and losing.

Write one paragraph about why the current way fails you. Ask for something that fixes that specific failure, with one easy way to put things in. Use it for two days before you ask for a single change.

If you want help deciding which parts of your business are worth handing over, that is what I do with teams. Have a look at how I work.